Open sbonner0 opened 2 years ago
Great help!
The code doesn't rely on OpenMP as a hard constraint. The extension loader in TorchDrug uses these lines to detect whether OpenMP is available. If OpenMP is not available, it will fall back to the native parallel of PyTorch. I have no idea what the speed difference is, but PyTorch recommends to use OpenMP for CPU parallel. If you only use Mac for debug purpose, I feel OpenMP is not a must.
I am not a Mac user, so please correct me if I am wrong.
Thanks!
I agree that OpenMP shouldn't be a hard constraint, however I still needed to take the steps detailed in the PR to even be able to install with on MacOS at all. I'll create a new venv and test if this is still the case.
Adding a few tips I needed to get this to run on an Intel based MacOS install.
I haven't tested on my M1 machine yet but will update further if any extra steps are needed.